Historically, data processing engines have been physically and conceptually separated from the memory which stores the data and program commands. As processor speed has increased over time, there has been a continuous press for larger memories and faster access. Recent advances in processor speed have caused system bottlenecks in access to memory. This restriction is critical because delays in obtaining instructions or data may cause significant processor wait time, resulting in loss of valuable processing time.
Various approaches have been taken to solve these concerns. Generally, the solutions include using various types of memory which have different attributes. For example, it is common to use a relatively small amount of fast, and typically expensive, memory directly associated with the processor units, typically called cache memory. Additionally, larger capacity, but generally slower, memory such as DRAM or SRAM is associated with the CPU. This intermediate memory is often large enough for a small number of current applications, but not large enough to hold all system programs and data. Mass storage memory, which is ordinary very large, but relatively inexpensive, is relatively slow. While advances have been continually made in improving the size and speed of all types of memory, and generally reducing the cost per bit of memory, there remains a substantial need especially to serve yet faster processors.
For the last 20 years most mass storage devices have utilized a rotating memory medium. Magnetic media have been used for both "floppy" (flexible) disks or "hard" disk drives. Information is stored by the presence or absence of magnetization at defined physical locations on the disk. Ordinarily, magnetic media are "read-write" memories in that the memory may be both written to and read from by the system. Data is written to or read from the disk by heads placed close to the surface of the disk.
A more recent development in rotating mass storage media are the optical media. Compact disks are read only memory in which the presence or absence of physical deformations in the disk indicates the data. The information is read by use of a focused laser beam, in which the change in reflectance properties from the disk indicate the data states. Also in the optical realm are various optical memories which utilize magneto optic properties in the writing and reading of data. These disks are both read only, write once read many ("WORM") drives and multiple read-write memories. Generally, optical media have proved to have a larger storage capacity, but higher costs per bit and limited write ability, as compared with magnetic media.
Several proposals have been made for using polymers for electronic based molecular memories. For example, Hopfield, J. J., Onuchic, J. N. and Beratan, D. N., "A Molecular Shift Register", Science, 241, p. 817, 1988, discloses a polymer based shift register memory which incorporates charge transfer groups. Other workers have proposed an electronic based DNA memory (see Robinson et al, "The Design of a Biochip: A Self-Assembling Molecular-Scale Memory Device", Protein Engineering, 1:295-300 (1987)). In this case, DNA is used with electron conducting polymers for a molecular memory device. Both concepts for these molecular electronic memories do not provide a viable mechanism for inputting data (write) and for outputting data (read).
Molecular electronic memories have been particularly disappointing in their practical results. While proposals have been made, and minimal existence proofs performed, generally these systems have not been converted to commercial reality. Further, a specific deficiency of the system described above is that a sequential memory is typically substantially slower than a random access memory for use in most systems.
The optical memories described above suffer from the particular problem of requiring use of optical systems which are diffraction limited. This imposes size restrictions upon the minimum size of a data bit, thereby limiting memory density. This is an inherent limit in systems which store a single bit of data at a given physical memory location.
Further, in all optical memory systems described above, the information is stored on a bit-by-bit basis, such that only a single bit of data is obtained by accessing a giving physical location in memory. While word-wide memory access systems do exist, generally they store but a single bit of information at a given location, thereby requiring substantially the same amount of physical memory space whether accessed in a bit manner or word-wide manner.
While systems have generally increased in speed and storage density, and decreased in cost per bit, there remains a clear gap at present between processor speed and system requirements. See generally, "New Memory Architectures to Boost Performance", Tom R. Halfhill, Byte, Jul., 1993, pp 86 and 87. Despite the general desirability of memories which are faster, denser and cheaper per bit, and the specific critical need for mass memory which can meet the demands of modern day processor systems speed, no completely satisfactory solution has been advanced heretofore. The fundamental limitations on the currently existing paradigms cannot be overcome by evolutionary enhancements in those systems. This invention constitutes a new memory paradigm.